Input / output operations per second
Input / Output operations Per Second , or IOPS for short , is a computer benchmark specification of data carriers . It indicates how many input and output commands can be carried out per second. The information can be specified more precisely:
- Total IOPS
Average number of inputs / outputs per second - Read IOPS
Average number of issues per second - Write IOPS
Average number of entries per second
The larger the IOPS, the faster the disk. New storage media, such as an SSD , achieve up to 20,000–50,000 IOPS. SSDs with Phison PS3110-S10 controller are said to achieve up to 110,000 IOPS (as of May 2017). The performance of conventional hard drives depends on the number of revolutions and the associated electronics. S-ATA-II drives 7200 min -1 and an average access time of 5 to 6 ms can reach about 120 IOPS while FC HDDs min with 15k -1 can reach about 340 IOPS.
While hard disks deliver performance, the upstream RAID controllers limit this performance. An enterprise virtual array system achieves 210,000 IOPS (EVA8100). In order to achieve this performance over the long term, it requires at least 620 hard drives. If values higher than the sum of all hard disks are achieved, this can only be achieved through the use of the cache, i.e. H. of the main memory of the storage system. (In the Unix and Windows sectors, caching usually takes place in the servers, which means that low cache hit rates are achieved in the storage system.)
criticism
The measured IOPS values depend on the access pattern (sequential / random) as well as on cache and / or I / O scheduler settings of the hard disk controller or the operating system, which is why more than 120 IOPS can usually be achieved with S-ATA -II hard drives (see above) can be measured. Manufacturer values in particular are therefore hardly comparable with one another; The paper data also does not provide any information about its own practical usability. The value 1 IOPS is also undefined, which makes comparisons with values in MB / s impossible. The IOPS measure lacks any scientific relevance and is only meaningful as a relative measured variable in a defined environment.
The value can therefore be used sensibly within a data center, for example to find the best possible disk configuration for a server. It also helps here if application manufacturers provide an application-related test, as Oracle does for its database from version 11.1.
See also
Individual evidence
- ↑ Product announcement Kingston
- ↑ DS4000 Best Practices and Performance Tuning Guide . (PDF; 11.7 MB) IBM Redbook SG24-6363, p. 36